Fetch per-numa memory usage from the OS.

[kilobyte]

This change is 

[bratpiorka]

this is not strictly related to hot poc - could you extend print_memtier_memory() with this info?

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @kilobyte)

[maciekpac]

Maybe it would be a better idea to use CONFIG_NODES_SHIFT or some other #def ?
What about this code?

[maciekpac]

return without fclose(f)

[maciekpac]

some comments would be welcome, the code could be more readable

[maciekpac]

wouldn't simple '''break''' have the same effect?

[kilobyte]

The values are global, not sure if it's good to print them there. On the other hand, if we marked that as global...

Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @maciekpac and @PatKamin)

---

src/numus.c, line 7 at r1 (raw file):

Previously, maciekpac wrote…

Maybe it would be a better idea to use CONFIG_NODES_SHIFT or some other #def ?
What about this code?

CONFIG_NODES_SHIFT is a settable piece of kernel config, and thus can't be used outside of the kernel itself. I don't see a #define for the max in the kernel's uapi either.

---

src/numus.c, line 41 at r1 (raw file):

Previously, maciekpac wrote…

wouldn't simple '''break''' have the same effect?

Alas, we're in a deeper while.

---

src/numus.c, line 53 at r1 (raw file):

Previously, maciekpac wrote…

some comments would be welcome, the code could be more readable

I've added some.

---

src/numus.c, line 61 at r1 (raw file):

Previously, maciekpac wrote…

return without fclose(f)

Done.

[bratpiorka]

ok but otherwise this code wouldn't be used and tested at all

Reviewed 1 of 1 files at r2, all commit messages.
Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @maciekpac and @PatKamin)

[PatKamin]

Reviewed 1 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 6 unresolved discussions (waiting on @kilobyte and @maciekpac)

---

src/numus.c, line 7 at r2 (raw file):

#include <stdlib.h>

#define MAX_NODES 1024 /* kernel max */

Perhaps you could use numa_num_possible_nodes() from numa library.

---

src/numus.c, line 75 at r2 (raw file):

        }
        for (int i = 0; i < maxnode; i++)
            numamem[num[i]].used += count[i] * pagesize;

size_t += long * unsigned long long

[kilobyte]

Done.

Reviewable status: 1 of 4 files reviewed, 6 unresolved discussions (waiting on @bratpiorka, @maciekpac, and @PatKamin)

---

src/numus.c, line 7 at r2 (raw file):

Previously, PatKamin (Patryk Kaminski) wrote…

Perhaps you could use numa_num_possible_nodes() from numa library.

In this version, the struct is static thus I'm using a compile-time value (that matches kernel max) — but once we get to polishing this code for outside usage, that's exactly the function we should use. Thanks for researching this bit, I've put it in a comment so it's not forgotten.

---

src/numus.c, line 75 at r2 (raw file):

Previously, PatKamin (Patryk Kaminski) wrote…

size_t += long * unsigned long long

count[i] is size_t, pagesize is uint64_t, both unsigned. Am I missing something?
I've just changed pagesize to size_t for consistency, but that's the same effective type.

[PatKamin]

Reviewed 3 of 3 files at r3, all commit messages.
Reviewable status: all files reviewed, 6 unresolved discussions (waiting on @kilobyte and @maciekpac)

---

src/numus.c, line 75 at r2 (raw file):

Previously, kilobyte (Adam Borowski) wrote…

count[i] is size_t, pagesize is uint64_t, both unsigned. Am I missing something?
I've just changed pagesize to size_t for consistency, but that's the same effective type.

Indeed. I wrongly assumed that count[i] is long because you use atol() when assigning it's value.

---

src/numus.c, line 62 at r3 (raw file):

                    continue;
                num[maxnode] = node;
                count[maxnode] = atol(valp);

atol() return long whereas count[maxnode] is an ull.

[maciekpac]

Reviewed 1 of 2 files at r1, 3 of 3 files at r3, all commit messages.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @kilobyte and @PatKamin)

---

src/numus.c, line 62 at r3 (raw file):

Previously, PatKamin (Patryk Kaminski) wrote…

atol() return long whereas count[maxnode] is an ull.

up

[kilobyte]

Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @PatKamin)

---

src/numus.c, line 75 at r2 (raw file):

Previously, PatKamin (Patryk Kaminski) wrote…

Indeed. I wrongly assumed that count[i] is long because you use atol() when assigning it's value.

→ strtoul()

---

src/numus.c, line 62 at r3 (raw file):

Previously, maciekpac wrote…

up

Changed to unchecked strtoul().

[PatKamin]

Reviewed 2 of 2 files at r4, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @kilobyte)

[kilobyte]

Rebased.

[maciekpac]

Reviewed 1 of 2 files at r4, 2 of 2 files at r5, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @kilobyte)